fauxneticien

def signal2int(signal, sample_rate=16_000, seq_length=4):
  """
  Decode a sine wave signal back into digits
  """
  digits = []

  for start_index in range(0, sample_rate * seq_length, sample_rate):
    end_index = start_index + sample_rate - 1

    window_samples = signal[start_index:end_index]

fauxneticien

from functools import reduce

input = 1
f1 = lambda i: i+1
f2 = lambda i: i+2
f3 = lambda i: i+3

output = reduce(lambda i, f: f(i), [f1, f2, f3], input)

fauxneticien

import torch
import torchaudio

fakewav_80mins = torch.rand(1, 16_000 * 60 * 80)
torchaudio.save("tmp/80mins.wav", fakewav_80mins, 16_000)

import nemo.collections.asr as nemo_asr

asr_model = nemo_asr.models.EncDecCTCModelBPE.from_pretrained(model_name="nvidia/stt_en_fastconformer_ctc_large")
asr_model.eval()

fauxneticien

import torchaudio

import pandas as pd

from pathlib import Path
from tqdm import tqdm

llight10h =  torchaudio.datasets.LibriLightLimited(root="tmp/", subset="10h", download=True)

manifest_rows = []

fauxneticien

import torch
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader

import os
import torchaudio
from lhotse import CutSet, Fbank, FbankConfig
from lhotse.dataset import IterableDatasetWrapper, SpeechSynthesisDataset, DynamicBucketingSampler, OnTheFlyFeatures, make_worker_init_fn
from lhotse.recipes import download_librispeech, prepare_librispeech
from tqdm import tqdm

fauxneticien

import torch
from transformers import Wav2Vec2Model
from torchaudio.models.wav2vec2.utils import import_huggingface_model

hf_model = Wav2Vec2Model.from_pretrained("facebook/wav2vec2-xls-r-300m")
ta_model = import_huggingface_model(hf_model)

torch.save(ta_model.state_dict(), "tmp/wav2vec2-300m_xls-r.pt")

## Usage

fauxneticien

20230710_mini-librispeech/
├── dev
│   ├── cuts.000000.jsonl.gz
│   ├── cuts.000001.jsonl.gz
│   ├── cuts.000002.jsonl.gz
│   ├── cuts.000003.jsonl.gz
│   ├── cuts.000004.jsonl.gz
│   ├── cuts.000005.jsonl.gz
│   ├── fbank.000000.tar
│   ├── fbank.000001.tar

fauxneticien

my_instantiated_object:
  _target_: my_classes.my_abstract_class.MyAbstractClass
  my_variable: used Hydra YAML!

fauxneticien

# Modified version with diff (see history)

class TokenCollater:
    """Collate list of tokens

    Map sentences to integers. Sentences are padded to equal length.
    Beginning and end-of-sequence symbols can be added.
    Call .inverse(tokens_batch, tokens_lens) to reconstruct batch as string sentences.

    Example:

fauxneticien

#!/usr/bin/bash
# Basic range in for loop
for value in {1..8}
do
    echo "Starting backup $value"
    rclone copy -P --exclude checkpoint_last.pt /home/nay/w2v2-fairseq-pretrain/outputs/2023-04-09/08-12-57/checkpoints/ gdrive_slasr:asr/checkpoints/xls-r_cpt_malay-140h/checkpoints/35-ga16/
    echo "Sleeping for 1 hour"
    sleep 1h
done